Approximate Top-$k$ for Increased Parallelism
- URL: http://arxiv.org/abs/2412.04358v1
- Date: Thu, 05 Dec 2024 17:17:28 GMT
- Title: Approximate Top-$k$ for Increased Parallelism
- Authors: Oscar Key, Luka Ribar, Alberto Cattaneo, Luke Hudlass-Galley, Douglas Orr,
- Abstract summary: We present an evaluation of bucketed approximate top-$k$ algorithms.
By relaxing the requirement that the top-$k$ is exact, bucketed algorithms can dramatically increase the parallelism available.
We also release a fast bucketed top-$k$ implementation for PyTorch.
- Score: 1.2557921586915128
- License:
- Abstract: We present an evaluation of bucketed approximate top-$k$ algorithms. Computing top-$k$ exactly suffers from limited parallelism, because the $k$ largest values must be aggregated along the vector, thus is not well suited to computation on highly-parallel machine learning accelerators. By relaxing the requirement that the top-$k$ is exact, bucketed algorithms can dramatically increase the parallelism available by independently computing many smaller top-$k$ operations. We explore the design choices of this class of algorithms using both theoretical analysis and empirical evaluation on downstream tasks. Our motivating examples are sparsity algorithms for language models, which often use top-$k$ to select the most important parameters or activations. We also release a fast bucketed top-$k$ implementation for PyTorch.
Related papers
- Efficiently Learning One-Hidden-Layer ReLU Networks via Schur
Polynomials [50.90125395570797]
We study the problem of PAC learning a linear combination of $k$ ReLU activations under the standard Gaussian distribution on $mathbbRd$ with respect to the square loss.
Our main result is an efficient algorithm for this learning task with sample and computational complexity $(dk/epsilon)O(k)$, whereepsilon>0$ is the target accuracy.
arXiv Detail & Related papers (2023-07-24T14:37:22Z) - Provably Efficient Reinforcement Learning via Surprise Bound [66.15308700413814]
We propose a provably efficient reinforcement learning algorithm (both computationally and statistically) with general value function approximations.
Our algorithm achieves reasonable regret bounds when applied to both the linear setting and the sparse high-dimensional linear setting.
arXiv Detail & Related papers (2023-02-22T20:21:25Z) - A One-Sample Decentralized Proximal Algorithm for Non-Convex Stochastic
Composite Optimization [10.762749887051546]
We propose two-time scale algorithms: ProxDAS-A and Proxcal$DASA-GT.
Unlike prior work, our algorithms achieve comparable complexity without requiring large batch sizes, more complex per-it operations, or stronger assumptions.
arXiv Detail & Related papers (2023-02-20T05:16:18Z) - Robust Methods for High-Dimensional Linear Learning [0.0]
We propose statistically robust and computationally efficient linear learning methods in the high-dimensional batch setting.
We instantiate our framework on several applications including vanilla sparse, group-sparse and low-rank matrix recovery.
For vanilla $s$-sparsity, we are able to reach the $slog (d)/n$ rate under heavy-tails and $eta$-corruption.
arXiv Detail & Related papers (2022-08-10T17:00:41Z) - A Push-Relabel Based Additive Approximation for Optimal Transport [5.111364864495785]
Exact algorithms for computing Optimal Transport can be slow.
We introduce a new and very simple approach to find an $varepsilon$approximation of the OT distance.
Our algorithm achieves a near-optimal execution time of $O(n2/varepsilon2)$ for computing OT distance.
arXiv Detail & Related papers (2022-03-07T21:40:14Z) - Minimax Optimization with Smooth Algorithmic Adversaries [59.47122537182611]
We propose a new algorithm for the min-player against smooth algorithms deployed by an adversary.
Our algorithm is guaranteed to make monotonic progress having no limit cycles, and to find an appropriate number of gradient ascents.
arXiv Detail & Related papers (2021-06-02T22:03:36Z) - Fast Parallel Algorithms for Euclidean Minimum Spanning Tree and
Hierarchical Spatial Clustering [6.4805900740861]
We introduce a new notion of well-separation to reduce the work and space of our algorithm for HDBSCAN$*$.
We show that our algorithms are theoretically efficient: they have work (number of operations) matching their sequential counterparts, and polylogarithmic depth (parallel time)
Our experiments on large real-world and synthetic data sets using a 48-core machine show that our fastest algorithms outperform the best serial algorithms for the problems by 11.13--55.89x, and existing parallel algorithms by at least an order of magnitude.
arXiv Detail & Related papers (2021-04-02T16:05:00Z) - Online Model Selection for Reinforcement Learning with Function
Approximation [50.008542459050155]
We present a meta-algorithm that adapts to the optimal complexity with $tildeO(L5/6 T2/3)$ regret.
We also show that the meta-algorithm automatically admits significantly improved instance-dependent regret bounds.
arXiv Detail & Related papers (2020-11-19T10:00:54Z) - On Effective Parallelization of Monte Carlo Tree Search [51.15940034629022]
Monte Carlo Tree Search (MCTS) is computationally expensive as it requires a substantial number of rollouts to construct the search tree.
How to design effective parallel MCTS algorithms has not been systematically studied and remains poorly understood.
We demonstrate how proposed necessary conditions can be adopted to design more effective parallel MCTS algorithms.
arXiv Detail & Related papers (2020-06-15T21:36:00Z) - Ranking a set of objects: a graph based least-square approach [70.7866286425868]
We consider the problem of ranking $N$ objects starting from a set of noisy pairwise comparisons provided by a crowd of equal workers.
We propose a class of non-adaptive ranking algorithms that rely on a least-squares intrinsic optimization criterion for the estimation of qualities.
arXiv Detail & Related papers (2020-02-26T16:19:09Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.